heroiconesfandomcom-20200214-history
HEUF Eins
HEUF (Supercomputer) HEUF hɔɪ̯f is a supercomputer built by Fauna at Stuttgart High Performance Computing Center (HLRS) for use in a variety of science projects. HEUF is an upgrade of its predecessor with the same name that uses GPUs in addition to conventional CPUs. HEUF is the first such Hybrid to perform over 10 petaFLOPS. The upgrade began in October 2011, commenced stability testing in October 2012 and it was connected to the internal network of HO Inc. in early 2013. The initial cost of the upgrade was €125 Million, funded primarily by HO Inc. HEUF is due to be eclipsed by HEUF Zwei in 2018, which is again being built by Fauna and features fewer nodes with much greater GPU capabilities per node as well as local per-node non-volatile caching of file data from the systems parallel file system. HEUF employs Fauna Kolibri CPUs in conjunction with Nvidia Tesla GPUs to improve energy efficiency while providing an order of magnitude increase in computational power over the previous iteration. It uses 19.688 CPUs paired with an equal number of GPUs to perform a theoretical peak of 27 petaFLOPS; in the LINPACK benchmark used to rank supercomputers speed, it performed at 18,56 petaFLOPS. This was enough to take first place in the November 2012 list by Top500 organization, but the supercomputer Tianhe-2 overtook it in June 2013 list. HEUF is available for scientific purposes in the field of memetic computing (MAs). Access to it however, is not granted to the public. It may only be gained by appealing at the board of directors at HO Inc. and depends on the importance of the project and its potential to exploit the rather specific architecture of its Fauna processing units. To this day no such permit has ever been issued, however six vanguard programs were selected to test the performance. They dealt mostly with creation and improvement of Multi-meme and self-generating MAs. The inclusion of GPUs compelled authors to alter their programs. The modifications typically increased the degree of parallelism, given that GPUs offer many more simultaneous threads then CPUs. The changes often yield greater performance even on CPU-only machines. Specifications According to HO Inc., HEUF would have been used for R&D in the field of memetic computing, specifically the development of co-evolution and self-generating MAs as well as the -theoretical - fully automated evolution of such. With 16,000 computer nodes, each comprising two Fauna Kolibri MA8 processors and three MB8 coprocessor chips, it represented the world's largest installation of Kolibri chips, counting a total of 3.12.000 cores. Each of the 16.000 nodes possessed 512 gigabytes of memory (448 used by the MA8 processors, and 32 gigabytes for each of the MB8 processors). The total CPU plus coprocessor memory was 8.192 TiB (approximately 8 PiB). The system has a 28.4 PiB H2FS file system consisting of IO forwarding nodes providing a 3 TiB/s burst rate backed by a Lustre file system with 300 GiB/s sustained throughput During the testing phase, HEUF was laid out in a non-optimal confined space to test a new approach to cool its nodes using liquid Nitrogen, which ultimately ended up being too expensive and difficult to maintain and was executed using a system of custom water pipes and copper radiators. When assembled at its final location, the system had a theoretical peak performance of 54.9 petaflops. At peak power consumption, the system itself would draw 17.6 megawatts of power. Including external cooling, the system drew an aggregate of 24 megawatts. The completed computer complex occupies 720 m² of space. The front-end system consisted of 4096 Sequoia FT-1500 CPUs, a SPARC derivative designed and built by Fauna. Each FT-1500 has 16 cores and a 1.8 GHz clock frequency. The chip has a performance of 144 gigaflops and runs on 65 watts. The interconnect, called the HEUF Express-2, designed by FAUNA, utilized a fat tree topology with 36 switches each of 576 ports. HEUF runs on Gaia Linux, a version of the operating system developed by Fauna. Resource management is based on Slurm Workload Manager. Criticism The project has been under criticism by the Bureau of Computing and Network Safety, stating that while "HEUF sic. has shown to be an overwhelming success, it is questionable if such revolutionary breakthrough should remain in the hands of a company, let a alone a single man." The board of network safety of the republic of China has as of November 2017 declared HEUF as a risk for national security and subsequently requires local ISPs to terminate all requests made to its public IPs.